Image Structure from Monotonic Dual Graph Contraction

摘要: The qualitative structure of images is much like the landscapes. 'Critical points' a landscape are summits, immits, and saddle points. These points connected through special curves on surface landscape. new approach computes this basic an image or from neighborhood sampled grid by process called monotonic dual graph contraction (MDGC). vertices graphs store information about gray level height as attributes. Edges represent connecting vertices. MDGC successively removes non-extrema original while it preserves connectivity between extrema level, property expressing least difference when moving one extremum to another extremum. Since represents planar well defined. performs simplifications such that in all local maxima survive minima survive. Hence we call them 'maximum graph' 'minimum respectively. focus paper description hierarchy height. Monotonic properties preserved during process. implementation described experimental results discussed.